
Research Themes
Six themes, each with a named academic lead
What a DEIS research theme is
DEIS organises its work into six research themes. Each theme has a named academic lead, a defined scientific focus, and a route from research into tools, standards and industrial practice.
A research theme is a long-running scientific territory that DEIS owns, not a project that ends. It gathers the questions the group is trying to answer, the people who answer them, the methods and tools that result, and the industrial settings where they’re tested. Themes overlap on purpose: assurance methods from one feed the tools of another.
Group leads: Prof. Yiannis Papadopoulos, Lead · Dr Koorosh Aslansefat, Co-Lead
Lead · Dr Koorosh Aslansefat
Making machine learning components safe to rely on: detecting when a model has left the conditions it was validated for, and assuring it afterwards.
Focus: SafeML · OOD and distribution shift · interpretability · SafeLLM · AI assurance
Lead · Dr Zhibao Mian
Using AI in safety engineering itself, so that analysis, diagnosis and decision support scale to systems too large to assess by hand.
Focus: AI-assisted safety analysis · prediction · diagnostics · decision support
Lead · Dr Kay Atefi
Attacks that end in a safety failure, and the co-assurance needed when a security control and a safety argument constrain each other.
Focus: Cyber disturbance · adversarial inputs · data poisoning · spoofing · safety-security co-assurance
Lead · Dr David Parker
Generating safety analyses from system models instead of writing them by hand, and optimising designs against dependability objectives.
Focus: HiP-HOPS · Andromeda · FTA and FMEA · optimisation · synthesis with AI
Lead · Prof. Yiannis Papadopoulos
Vehicles, vessels and aircraft that decide for themselves, and the runtime assurance that has to hold when the operating situation was never anticipated.
Focus: Autonomous vehicles · maritime autonomy · UAV, ASV and AUV systems · runtime assurance
Led by the main tool authors, with Martin Walker where funding allows
The engineering that turns a method into something an industrial team can run, support and depend on.
Focus: HiP-HOPS · safety monitoring tools and EDDIs · demonstrators · industrial deployment
How the themes connect
The themes aren’t six separate groups. A model-based analysis method becomes a tool; the tool is applied to an autonomous vessel; the vessel raises a runtime assurance question; answering it needs a safety-security argument. Work moves across the boundaries in both directions.
Methods: themes 01, 03 and 04 produce the analysis and assurance methods.
Tools: themes 02 and 06 make them usable at industrial scale.
Application: theme 05 tests them where failure has real consequences.

Working on one of these problems?
We collaborate with industry on assurance of AI-enabled and autonomous systems, and supervise PhD research across all six themes.